Commit Graph

4 Commits

Author SHA1 Message Date
Vasyka 0e3f9e8bca feat: AI model selector + i18n nav labels (RU/EN) on new modules
AI model selector:
- AiAssistantService::MODEL_DEFAULTS and MODEL_OPTIONS const tables (3 picks per
  provider: Claude Opus 4.7 / Sonnet 4.6 / Haiku 4.5, OpenAI 4o / 4o-mini,
  Gemini 1.5 Pro / Flash). Default upgraded from Sonnet 4.5 → Sonnet 4.6.
- modelFor(provider, company?) resolves tenant override > global default.
- All 8 hardcoded model strings replaced with modelFor() across callClaude
  (chat with tool-use), callOpenAI, callGemini (chat), postClaude/postOpenAI/
  postGemini (single-shot), and OcrInvoiceService.
- Settings page adds 3 model selectors per provider with persistence at
  settings.ai.models.{claude,gpt,gemini}.

i18n nav labels:
- TireSet / Bodyshop / Subcontractor / SubcontractJob / PricingCoefficient /
  ShopCustomer resources: getNavigationLabel / getNavigationGroup /
  getModelLabel / getPluralModelLabel return __()-wrapped strings.
- 20 keys added to lang/ru.json and lang/en.json.

Tests (4 new): default model, tenant override wins, unknown provider falls
back to claude default, options dictionary contains each default key.

Full suite: 134 passed.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-06-03 06:23:21 +00:00
Vasyka dfb92bf5e2 feat: AI chat tool-use (Claude function calling)
The Asistent AI chat can now query the tenant DB directly through Claude
tool use. AiToolExecutor exposes 5 read-only tools (search_clients,
get_vehicle, find_parts, recent_workorders, low_stock_parts) all scoped
to the current tenant via BelongsToTenant.

AiAssistantService::callClaude loops on stop_reason=tool_use up to 5 rounds:
- normalize message history to content blocks
- send `tools` definitions + messages to Anthropic API
- on tool_use: execute each tool, append tool_results as user turn, recall
- on end_turn: collect text + cumulative token counts + tool-call audit in
  AiMessage.meta.tools

Single-shot helpers (suggestDiagnosis, suggestPrice, vinRecommendations,
suggestParts) are unchanged — only the conversational chat gets tool-use.

Tests (3 new):
- two-round tool_use → execute → final text; verify 5 tools sent both rounds;
  cumulative tokens
- executor finds part by brand
- unknown tool name returns error blob

Full suite: 109 passed.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-06-02 19:36:07 +00:00
Vasyka 1ff888131f Stage 16 — AI Layer: VIN decoder + diagnostic / parts / price helpers
VinDecoder (deterministic, no API):
- ISO 3779/3780 parsing: WMI manufacturer (~60 brands), year (cyclical with
  post-2010 disambiguation via position 7), region, plant, NA checksum
- Strip non-VIN chars, accept dashes/spaces, reject I/O/Q per spec

AiAssistantService:
- Refactored provider HTTP into postClaude/postOpenAI/postGemini so both
  chat history and one-shot calls share the same transport
- singleShot(system, userPrompt, provider?) for fire-and-forget calls
- 4 specialized helpers with tight prompts:
  - suggestDiagnosis(WO) — diagnostician based on complaint + VIN info
  - suggestParts(WO, task) — OEM parts list for an operation
  - suggestPrice(Part) — markup recommendation with justification
  - vinRecommendations(vin, mileage) — scheduled maintenance from decoded VIN
- monthlyUsage() — token spend MTD by provider

Filament:
- VehicleResource: "Decode VIN" + "AI: recomandări" actions
- WorkOrderResource Edit: "AI: sugerează diagnostic" header action
- PartResource: "AI: preț recomandat" action
- Shared views: filament.tenant.ai-reply, filament.tenant.vin-decode
- AiAssistant page shows monthly token usage banner

Tests (13 new):
- 8 VinDecoder unit tests with real VIN samples (Honda 2003, VW 1999, Audi
  2014, Dacia, unknown WMI, lowercase/dashes, forbidden chars)
- 5 AiHelpers feature tests with Http::fake covering all providers + no-key
  fallback + token usage aggregation

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-27 20:24:09 +00:00
Vasyka 976c0f03e3 AI Assistant — multi-provider chat (Claude / GPT / Gemini)
Schema:
- ai_chats: company_id, user_id, title, provider; index pe activitate
- ai_messages: role (system/user/assistant), content, meta JSON (tokens, latency, model)

Service AiAssistantService (multi-provider):
- ask($chat, $message): persistă mesajul user, build system prompt cu context
  tenant (statistici clienți/mașini/cereri/datorii), apelează API-ul providerului,
  persistă răspunsul cu meta (tokens, latency)
- callClaude: api.anthropic.com/v1/messages cu claude-sonnet-4-5
- callOpenAI: api.openai.com/v1/chat/completions cu gpt-4o-mini
- callGemini: generativelanguage.googleapis.com cu gemini-1.5-flash
- Try/catch pe toate; eroare devine mesaj asistent fără să crape

System prompt include:
- Numele și orașul companiei
- Statistici curente (clienți, mașini, cereri noi, fișe active, datorii)
- Limita stricta: NU inventează date

Custom Filament Page /app/ai-assistant (group Analiză):
- Sidebar stâng: listă conversații (last 20), buton 'Nouă' + delete cu confirm
- Main: bubble chat (user dreapta albastru, asistent stânga gri)
- Meta jos pe răspuns: provider · latency · tokens
- Empty state friendly cu instrucțiuni configurare
- Loading indicator (3 dots animate) când AI răspunde
- Auto-scroll la mesaj nou
- Enter trimite, Shift+Enter newline
- Auto-titlu chat din primul mesaj user (60 chars)

Settings page extins cu secțiune 'Asistent AI':
- Provider implicit (claude/gpt/gemini)
- 3 chei API (password fields, revealable)
- Key-urile salvate în companies.settings.ai (per tenant, izolat)
2026-05-07 14:50:56 +00:00